io_uring: fix early sqd_list removal sqpoll hangs
authorPavel Begunkov <asml.silence@gmail.com>
Tue, 13 Apr 2021 10:43:00 +0000 (11:43 +0100)
committerJens Axboe <axboe@kernel.dk>
Wed, 14 Apr 2021 19:07:27 +0000 (13:07 -0600)
commitc7d95613c7d6e003969722a290397b8271bdad17
tree5eddfc6c45f59cea3351e2b0e3ffc3adb5cb2d1e
parentc60eb049f4a19ddddcd3ee97a9c79ab8066a6a03
io_uring: fix early sqd_list removal sqpoll hangs

[  245.463317] INFO: task iou-sqp-1374:1377 blocked for more than 122 seconds.
[  245.463334] task:iou-sqp-1374    state:D flags:0x00004000
[  245.463345] Call Trace:
[  245.463352]  __schedule+0x36b/0x950
[  245.463376]  schedule+0x68/0xe0
[  245.463385]  __io_uring_cancel+0xfb/0x1a0
[  245.463407]  do_exit+0xc0/0xb40
[  245.463423]  io_sq_thread+0x49b/0x710
[  245.463445]  ret_from_fork+0x22/0x30

It happens when sqpoll forgot to run park_task_work and goes to exit,
then exiting user may remove ctx from sqd_list, and so corresponding
io_sq_thread() -> io_uring_cancel_sqpoll() won't be executed. Hopefully
it just stucks in do_exit() in this case.

Fixes: dbe1bdbb39db ("io_uring: handle signals for IO threads like a normal thread")
Reported-by: Joakim Hassila <joj@mac.com>
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/io_uring.c